Evangelista Francesco A, Simmonett Andrew C, Allen Wesley D, Schaefer Henry F, Gauss Jürgen
Center for Computational Chemistry, University of Georgia, Athens, Georgia 30602, USA.
J Chem Phys. 2008 Mar 28;128(12):124104. doi: 10.1063/1.2834927.
We report the first implementation with correct scaling of the Mukherjee multireference coupled cluster method with singles, doubles, and approximate iterative triples (Mk-MRCCSDT-n, n=1a,1b,2,3) as well as full triples (Mk-MRCCSDT). These methods were applied to the classic H4, P4, BeH(2), and H8 model systems to assess the ability of the Mk-MRCCSDT-n schemes to accurately account for triple excitations. In all model systems the inclusion of triples via the various Mk-MRCCSDT-n approaches greatly reduces the nonparallelism error (NPE) and the mean nonparallelism derivative diagnostics for the potential energy curves, recovering between 59% and 73% of the full triples effect on average. The most complete triples approximation, Mk-MRCCSDT-3, exhibits the best average performance, reducing the mean NPE to below 0.6 mE(h), compared to 1.4 mE(h) for Mk-MRCCSD. Both linear and quadratic truncations of the Mk-MRCC triples coupling terms are viable simplifications producing no significant errors. If the off-diagonal parts of the occupied-occupied and virtual-virtual blocks of the Fock matrices are ignored, the storage of the triples amplitudes is no longer required for the Mk-MRCCSDT-n methods introduced here. This proves to be an effective approximation that gives results almost indistinguishable from those derived from full consideration of the Fock matrices.
我们报告了首次实现的具有正确缩放比例的穆克吉多参考耦合簇方法,该方法包括单双激发以及近似迭代三激发(Mk-MRCCSDT-n,n = 1a、1b、2、3)和全三激发(Mk-MRCCSDT)。这些方法被应用于经典的H4、P4、BeH(2)和H8模型系统,以评估Mk-MRCCSDT-n方案准确考虑三激发的能力。在所有模型系统中,通过各种Mk-MRCCSDT-n方法包含三激发极大地降低了非平行误差(NPE)以及势能曲线的平均非平行导数诊断,平均恢复了全三激发效应的59%至73%。最完整的三激发近似,即Mk-MRCCSDT-3,表现出最佳的平均性能,将平均NPE降低到0.6 mE(h)以下,而Mk-MRCCSD的平均NPE为1.4 mE(h)。Mk-MRCC三激发耦合项的线性和二次截断都是可行的简化方法,不会产生显著误差。如果忽略福克矩阵占据-占据和虚-虚块的非对角部分,则此处介绍的Mk-MRCCSDT-n方法不再需要存储三激发振幅。事实证明,这是一种有效的近似方法,其结果与完全考虑福克矩阵得出的结果几乎无法区分。